import numpy as np
import pandas as pd
import matplotlib
matplotlib.use('Qt5Agg')
%matplotlib inline

import matplotlib.pyplot as plt

import seaborn as sns

import plotly.express as px


from pandas_profiling import ProfileReport

from dataprep.eda import create_report


from sklearn.model_selection import train_test_split
from sklearn.metrics import confusion_matrix
from sklearn.metrics import classification_report


import catboost
from catboost import *
import shap
shap.initjs()

from IPython.core.display import display, HTML
display(HTML("<style>.container { width:80% !important; }</style>"))

import warnings
warnings.filterwarnings('ignore')

from catboost import Pool, cv
from catboost import CatBoostClassifier, Pool, metrics, cv

Load Data

folder_name = '../data/'
file_name = '../data/click_data.txt'
data_df = pd.read_csv(file_name, delimiter='\t', parse_dates=['LogEntryTime'])
data_df['day_of_week'] = data_df['LogEntryTime'].dt.day_name()

Explore

create_report(data_df).show_browser()

What is the CTR by Device Type, OSFamily, and Location?

for f in ['DeviceType', 'OSFamily', 'Location']:
    _ = data_df.groupby(f)['Click'].mean()
    fig = px.bar(_, x="Click", )
    fig.update_layout(yaxis={'categoryorder':'total descending'})
    fig.show()

What are the top 10 sites with highest clicks?

_ = data_df.groupby('site')['Click'].sum().sort_values(ascending=False)[:10]
fig = px.bar(_, x="Click", orientation='h')
fig.update_layout(barmode='stack', xaxis={'categoryorder':'total descending'})

data_df['_site'] = data_df.site
data_df.loc[data_df['_site'].value_counts()[data_df['_site']].values < 2000, '_site'] = "Others"

for f in ['DeviceType', 'OSFamily', 'AdFormat', 'Location', '_site']:
    print(f)
    g = sns.catplot(data=data_df, x=f, col='Click', kind='count')
    plt.show()

DeviceType

OSFamily

AdFormat

Location

_site

What are the mean winningPriceCPMInBucks values for each day (2019-02-01 to 2019-02-07) and day of the week included in the data?

#(2019-02-01 to 2019-02-07)
_ = data_df.set_index('LogEntryTime').loc['2019-02-01':'2019-02-07']
_.groupby('day_of_week')['WinningPriceCPMInBucks'].mean()

day_of_week
Saturday    3.597775
Name: WinningPriceCPMInBucks, dtype: float64

Which features or mix of features out of DeviceType, OSFamily, AdFormat, Location, Site, and Day of the week are the most predictive for generating a click? Please explain your work

cat_features = ['DeviceType', 'OSFamily', 'AdFormat', 'Location', '_site', 'day_of_week']

X = data_df[cat_features]
y = data_df['Click']

X = X.fillna('None')
X_train, X_test, y_train, y_test = train_test_split(X, y,  test_size=0.25, random_state=45, stratify=y)
X.head()

	DeviceType	OSFamily	AdFormat	Location	_site	day_of_week
0	Mobile	iOS	300x250	???	Others	Tuesday
1	PC	OSX	728x90	1	Others	Thursday
2	PC	Windows	300x250	???	Others	Wednesday
3	PC	Windows	728x90	???	Others	Thursday
4	PC	OSX	300x250	3	quizlet.com	Sunday

model = CatBoostClassifier(iterations=100, cat_features=cat_features, learning_rate=0.1, random_seed=45)
model.fit(X_train, y_train, verbose=False, plot=False)

<catboost.core.CatBoostClassifier at 0x7ffe9aef0df0>

importances = model.feature_importances_
indices = np.argsort(importances)
features = X_train.columns
plt.title('Feature Importances')
plt.barh(range(len(indices)), importances[indices], color='g', align='center')
plt.yticks(range(len(indices)), [features[i] for i in indices])
plt.xlabel('Relative Importance')
plt.show()

explainer = shap.TreeExplainer(model)
shap_values = explainer.shap_values(Pool(X, y, cat_features=cat_features))

# summarize the effects of all the features
shap.summary_plot(shap_values, X)

Model

pool = Pool(data=X_train, label=y_train, cat_features=cat_features)

model = CatBoostClassifier(iterations= 100, 
                           loss_function = "Logloss", 
                           eval_metric='Recall', 
                           scale_pos_weight=5, 
                           learning_rate=0.1, 
                           verbose=False)

model.fit(
    X_train, 
    y_train,
    cat_features=cat_features,
    eval_set=(X_test, y_test),
    plot=True
)

<catboost.core.CatBoostClassifier at 0x7ffea2479340>

confusion_matrix(y_test, model.predict(X_test))

array([[25624, 90996],
       [  191, 31373]])

print(classification_report(y_test, model.predict(X_test)))

              precision    recall  f1-score   support

           0       0.99      0.22      0.36    116620
           1       0.26      0.99      0.41     31564

    accuracy                           0.38    148184
   macro avg       0.62      0.61      0.38    148184
weighted avg       0.84      0.38      0.37    148184

model.plot_tree(
    tree_idx=0,
    pool=pool
)